Research On Age Estimation Of Face Image Based On Deep Learning And Skin Texture Detection

As a key information dimension,age has important market potential value in the fields of public safety,business publishment and human-computer interaction.The age estimation based on face images is to use computer vision related technology to explore the mapping relationship between age distribution and changes in face images.By establishing a corresponding model,the algorithm can determine the age range and even the exact age of the individual through the provided image,thereby achieving non-contact age prediction.As a chronological biological feature,when the age range is similar,the texture features reflected on the human face have a certain similarity.In response to these challenges,this paper proposes an algorithm for age prediction of face images.The specific work includes:While ensuring accuracy,reducing the deviation of erroneous estimates,analyzing the most sensitive and representative skin characteristics of age changes as the detection target.Using convolutional neural networks to extract deep semantic features of human faces,and based on the target detection model to achieve skin quality detection to establish an innovative fusion prediction model.In view of the particularity of the skin detection task,a small target optimization method that incorporates dense anchor frames and combines channel weights and attention mechanisms is proposed,and the captured feature information is integrated for age regression.Experiments were conducted on the FG-NET database and the MORPH II dataset,and the interval labels of the GROUP dataset were used to verify the unrestricted adaptability of the algorithm.In order to adapt to the particularity of facial features changing at an individual's age,an age feature separation algorithm based on adversarial training is proposed.With the idea of generative confrontation,it is possible to decorrelate the identity information dimension and age feature dimension of the face image,and strip out the age information feature vector hidden in the face image.Finally,MAE values of 2.97 and 3.30 were obtained on the FG-NET database and MORPH-II dataset,respectively.The experimental results show that,compared with a single classification or regression task,the skin quality detection module combined with the individual age separation model can obtain higher accuracy and better robustness in age estimation.